Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75054.9431368965,
-      "required_pages_per_second_upper_estimate": 0.00650557224609193
+      "required_token_positions_per_second": 75571.69075210567,
+      "required_pages_per_second_upper_estimate": 0.006550362619693404
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9381.867892112063,
-      "required_pages_per_second_upper_estimate": 0.003742567373588664
+      "required_token_positions_per_second": 9446.461344013209,
+      "required_pages_per_second_upper_estimate": 0.0037683346673101996
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18763.735784224125,
-      "required_pages_per_second_upper_estimate": 0.00434066004320922
+      "required_token_positions_per_second": 18892.922688026418,
+      "required_pages_per_second_upper_estimate": 0.004370545159792027
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28145.60367633619,
-      "required_pages_per_second_upper_estimate": 0.0017139195953188129
+      "required_token_positions_per_second": 28339.384032039627,
+      "required_pages_per_second_upper_estimate": 0.0017257198022941993
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28145.60367633619,
-      "required_pages_per_second_upper_estimate": 0.002166726483021927
+      "required_token_positions_per_second": 28339.384032039627,
+      "required_pages_per_second_upper_estimate": 0.0021816442312223346
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28145.60367633619,
-      "required_pages_per_second_upper_estimate": 0.34947420038412397
+      "required_token_positions_per_second": 28339.384032039627,
+      "required_pages_per_second_upper_estimate": 0.3518803038608295
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.144230393796704,
-    "producer_p95_seconds": 0.39276865400461247,
-    "gpu_update_seconds": 0.9823203764943739,
-    "producer_batches_per_second": 6.933351380912976,
-    "gpu_batches_per_second": 1.0179978181545208,
-    "producer_headroom": 6.810772338866222,
+    "producer_batch_seconds": 0.11853516871843794,
+    "producer_p95_seconds": 0.39604436499939766,
+    "gpu_update_seconds": 0.9756034206227641,
+    "producer_batches_per_second": 8.436314815355315,
+    "gpu_batches_per_second": 1.0250066562514333,
+    "producer_headroom": 8.230497591311147,
     "sustainable": true,
-    "buffer_seconds": 60.90386334265118,
+    "buffer_seconds": 60.48741207861137,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469244.1977292965,
-    "upload_bytes_per_second": 5759103.241921413,
+    "generation_bytes_per_second": 453232.10291434894,
+    "upload_bytes_per_second": 5725874.910130629,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.68888138684436,
+    "estimated_full_shards_per_hour": 97.253058582047,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.2221524419983325,
-      "bytes": 6743871,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.1651974790001987,
       "bytes": 11511318,
@@ -562,6 +556,12 @@
       "bytes": 2483869,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.5748863410044578,
+      "bytes": 6345410,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 229700, "loss": 0.6110566407442093, "lr": 0.0003, "grad_norm": 0.15732614767708517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33621456149, "tokens_per_second": 147668.8134239256, "data_wait_s": 1.0698000551201403e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 229800, "loss": 0.5975795947015285, "lr": 0.0003, "grad_norm": 0.14342664525978513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33636017226, "tokens_per_second": 147616.1027512308, "data_wait_s": 8.350005373358727e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 229900, "loss": 0.5572257284075022, "lr": 0.0003, "grad_norm": 0.1450757103888371, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33650707927, "tokens_per_second": 149624.6344856014, "data_wait_s": 1.2054995750077069e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 230000, "loss": 0.6125554777681828, "lr": 0.0003, "grad_norm": 0.14557966935096453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33665466237, "tokens_per_second": 149953.93094908562, "data_wait_s": 1.871899439720437e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:76745cfb0f380dc7a1fe45890e7c4ce9088918444114344bee985a7a89db21b4
-size 744045
+oid sha256:db29c7907daf959b4cb154f421f0b06ca7afa8f445a409773ee034123004044e
+size 746163